[pdal] FieldChooserFilter

[Michael P. Gerlek]

> Does this filter result in the PointBuffer's data being copied/composed?
The
> scaling filter currently has to copy along all of the other fields' data,
and this is
> a performance hit because the PointBuffer's composition has changed
> (dropped int-based XYZ and added double-based XYZ or vice versa).

Way back when, I had the idea of a field that could be marked "unused" for
exactly this reason.  I wanted the writer to start off by creating and
passing down a schema'd buffer that (in your case) would have looked like
this:

  - int x
  - int y
  - int z
  - double x
  - double y
  - double z

The stages would only play with the fields they understood, but the buffer
was always the superset of all possible fields used by all stages.  The
buffer would waste space, but not much of a worry given that we've got a
good chunking system in place.

Maybe we should think about that idea again.

-mpg